The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system utilities including ffmpeg, ffprobe, yt-dlp, and ASR tools like whisper and agent-reach via subprocess.run. These tools are necessary for audio processing and video metadata extraction. Arguments are passed as lists to prevent shell injection, and the skill performs validation on user-supplied URLs.
[EXTERNAL_DOWNLOADS]: The skill retrieves video data and transcripts from supported platforms (YouTube, Bilibili, Xiaoyuzhou, Douyin, and Xiaohongshu). It also optionally interfaces with the Groq API for transcription services. These external connections are essential for its primary transcription workflow.
[PROMPT_INJECTION]: The skill processes transcription data from external sources which could potentially contain malicious instructions intended to influence the agent's behavior during summarization or translation (Indirect Prompt Injection).
Ingestion points: Data is ingested in transcribe_url within scripts/space_video_transcript.py after being fetched by platforms.py.
Boundary markers: The output format in scripts/formats.py provides basic metadata headers but does not include explicit delimiters to isolate the transcript content from agent instructions.
Capability inventory: The skill has the capability to write files, perform network requests to the Groq API, and execute local subprocesses.
Sanitization: The clean_text function in scripts/formats.py removes HTML tags and zero-width characters but does not filter or sanitize natural language instructions within the text content.

space-video-transcript